Developed a feature-rich inference service for the UPTAC-KomSai-v2/ISKOLMATE repository, delivering an end-to-end image processing pipeline using Python, Flask, and TensorFlow. The work centered on building a scalable Flask application that supports image uploads and predictions, orchestrating parallel loading of multiple Keras models through threading to enhance throughput. Implemented ensemble voting mechanisms to aggregate predictions for improved accuracy, and incorporated multiple color-space transformations for robust image preprocessing. The solution focused on reliability and scalability, enabling faster and more accurate image analysis. No major bugs were reported, reflecting a strong emphasis on quality and maintainability throughout the development process.